It could be a monitor issue. I have noticed a lot of people who have this problem seem to have Dell monitors so I'm wondering if Vega drivers have some kind of issue communicating with the monitor during idle.
Also wanted to note that doing a clean Windows 10 install didn't work for me, and locking HBM at any state hasn't worked either.
Although since 19.3.3 the crashes seem to be less frequent. They only seem to happen on the Windows login screen now if I leave it to idle there. The crashes don't seem to happen if I'm logged into my account.
It's an underlying hardware design fault that occurs mostly independently of total display bandwidth, but instead it's based on SoC voltage and signal drain.
The data stream sent to displays is sent directly out of the SoC fabric links, same as they are for the ryzen APUs (hence sharing the same pins for PCIe and display output). However, because the SoC was incorrectly wired to the core voltage rail, as opposed to a dedicated SoC rail that it's supposed to have, the voltage variance in the core causes the SoC to collapse from irregular drain.
The 14nm HBM2 vega series is the only series with the fault. The VII that you got doesn't exhibit the issue because they fixed the SoC problem by giving it it's dedicated rail. Hence the double-controller situation, that buildzoid from GamersNexus wasn't actually aware was needed (hence my above request to tell them about it).
So there is no hope they will ever fix this via driver or something? We are stuck with having to manually control HBM2 frequency?
BTW, are these settings still keep resetting from time to time with the newest drivers? I haven't updated for a few versions now as I was afraid of changing anything once I found a stable settings. But I have to check the settings every frikkin' day which is kind of annoying.
AMD's official response is to return the card/s, and yes, you have to re-configure wattman every update. They did however improve wattman in the latest versions as to not forget settings so easily.
Really? So we are fully eligible for return and refund? Do 3rd party manufacturers like Sapphire honor that?
Yes you can return your GPU to the retailer if it is under Warranty and you are the original purchaser.
Provided the GPU is in good physical condition and not been opened you should have no problems at all.
You will likely need to do some technical debug and show evidence of the problem.
Obviously if you damaged the card by smashing it against a wall in frustration or jumping on it they will not accept a return.
I am thinking of doing that with my RX Vega 64 Liquid.
Bought it end 2018. not used for months as it is so unstable.
They can either off you a replacement of the same model or similar.
Unlikely you would get a refund unless you return it within a short time of purchase depending where you are.
Good luck with your return.
Well what good another card with the same problem does? And the "similar" model... well it's a design error shared by all Vegas. And I wouldn't go for Nvidia card. So the only option would be to get a full refund and buy... well, there is that problem as well. Buy what? I don't need the power of Radeon VII, certainly not for an additional 300eur, Vega is a sweet spot for me. So yeah, we need another GPU manufacturer for sure.
I forgot to mention that you might have to go to teh manufacturer suppot and open a Support Ticket and het an RMA approved.
RE: And I wouldn't go for Nvidia card.
Why. Do you think AMD treat you better?
I purchased an RTX 2080 end last year because the RX Vega 64 Liquid just kept on crashing.
The RTX2080 cost about the same as the RX Vega 64 Liquid.
It is a much better and faster GPU.
I have run it every day with no problem.
Nothing like that, I would gladly buy Nvidia but their drivers cause an additional real-time audio latency, and I use this PC as a digital audio workstation so the low latency is my priority. It is a very low amount though, so I might go nvidia in the end.
Thx for the replies btw.
Well, you could get polaris if you don't need that much performance, or just go with the VII and set the TDP low. Or of course wait and see what navi brings.
RMA wise, it depends on the country and the retailer's sensibility. I had my v64 for over a year before I finally RMA'ed it, and while they weren't able to directly replicate the bug they refunded it anyway, but being Australia this is generally to be expected. Europe and the UK do have similar laws however.